RL for VLMs

RL: Reinforcement Learning · VLM: Vision-Language Model

Follow this topic Ask about this topic

Momentum

36 papers in the last four weeks, up 227% on the four weeks before. 0.4% of all new papers.

Jul 13Week of Sep 28

Latest papers 287

Sep 21, 2026cs.RO

Imagine-RL: Residual-Confidence-Guided Cross-Attention for World-Model-Augmented VLA Reinforcement Learning

Reliable action evaluation in contact-rich manipulation requires looking beyond the current observation to future visual and contact consequences. Existing noise-space reinforcement learning efficiently steers a frozen Vision-Language-Action (VLA) policy, but its critics largely ignore these consequences. We present Imagine-RL, which augments noise-space VLA post-training with action-conditioned visual-torque imagination. For each candidate action chunk, a frozen visual-torque latent world model (VTLWM) autoregressively predicts compact future representations without pixel reconstruction. A current image-state-action query attends to observed histories and predicted futures, while previous-window prediction residuals provide token-wise confidence priors that suppress unreliable future tokens. By combining current evidence with predicted consequences, the action critic better evaluates candidate actions and supervises the actor, while the VLA and VTLWM remain frozen. Across four real-robot tasks with 50 evaluation trials per task, Imagine-RL uses only 100 RL trajectories and improves the average success rate by (23.6%) over DSRL and by (60%) over VLA baselines.

Kejia Hu, Wentong Zhai, Bo Zhao +1

Department of Electric Engineering, Korea Advanced Institute of Science and Technology · University Of Science and Technology Beijing · Shanghai Jiao Tong University +1

Robotic Manipulation Vision-Language-Action Models RL for VLMs Model-Based RL

Sep 17, 2026cs.CV

Region-Level Policy Optimization for Fine-grained MLLM Perception

Fine-grained visual perception in MLLMs is commonly improved by raising the resolution, but the added visual tokens inflate vision-encoding and language-model prefilling costs. We show that the two operations underlying fine-grained perception, localizing the region of interest (RoI) and recognizing its content, have different resolution requirements. In a controlled diagnostic, localization tolerates roughly 3 to 4 times stronger token compression than recognition, which motivates localizing from a coarse view and concentrating resolution on the selected evidence. Decoding coordinates with the MLLM can be trained end-to-end from answers, but costs a full model pass per query and depends on grounding ability. A lightweight proposal network distilled from the model's attention is fast, but inherits the noise of its attention targets. The RoI from the proposal network reaches the answer through a discrete region choice, so its faithfulness to the answer cannot supervise the network. We therefore optimize the proposal network with region-level reinforcement learning, which we call Vision-RL2. It treats coherent regions as actions, and a frozen MLLM reader scores each one by how its removal changes the answer likelihood. Complementary subtractive and additive objectives suppress distracting proposals and recover missing evidence, updating only the predictor without region annotations, response sampling, or reasoning trajectories. The refined proposal further enables a sparse encoding that magnifies evidence and excludes background tokens. Across six fine-grained benchmarks and four MLLM backbones, Vision-RL2 improves accuracy over the base model at every token budget and surpasses its largest-budget accuracy with about 4 times fewer visual tokens. Code is available at https://github.com/YuHengsss/VisionRL2 .

Yuheng Shi, Xiaohuan Pei, Minjing Dong +1

University of Sydney · City University of Hong Kong

Efficient VLM Inference Large Vision-Language Models Fine-Grained Visual Perception Visual Token Pruning RL for VLMs

Sep 16, 2026cs.RO

GroundingVLN: Reasoning and Acting with Grounding for Vision-Language Navigation

Although vision-language models (VLMs) possess strong visual understanding and reasoning capabilities, existing vision-and-language navigation (VLN) agents struggle to connect semantic reasoning with spatial execution. Two coupled gaps remain in this connection, as intermediate reasoning is not explicitly anchored to visual evidence and high-level decisions lack precise spatial goals to guide low-level motion. Cognitive science suggests that human navigation bridges these levels hierarchically by anchoring cognition to relevant landmarks and guiding locomotion toward spatial goals. Motivated by this principle, we propose GroundingVLN, which uses visual grounding as a shared interface between reasoning and action. GroundingVLN first reasons with grounding by anchoring task-relevant visual evidence to precise image locations throughout structured reasoning. It then acts through grounding by predicting a progress-aligned pixel goal that a geometric planner translates into primitive actions. To learn these capabilities, we construct GroundingCOTVLN-188K, a dataset of temporally aligned grounded reasoning traces, and introduce Grounded and Execution-Aware Reinforcement Learning (GEAR), which aligns grounded reasoning and spatial decisions with downstream execution. Experiments demonstrate that GroundingVLN achieves state-of-the-art performance (69.9% SR on R2R-CE and 75.1% SR on RxR-CE) with high sample efficiency, using just 0.9% as much training data as the strongest baseline. It also generalizes strongly across datasets, attaining 59.9% SR on RxR-CE when trained solely on R2R, a gain of 20.1% over the strongest baseline. Code and models will be released after review.

Kailing Li, Yu Han, Tianwen Qian +4

School of Computer Science and Technology, East China Normal University · NeoteAI · King Abdullah University of Science and Technology

Vision-Language Navigation Vision-Language Grounding Visually Grounded Reasoning RL for VLMs Reasoning for VLA Models

Sep 16, 2026cs.AI

Anchoring What Matters: A Dual-Level Learning Framework for Visually-Grounded Multimodal Reasoning

Reinforcement learning with verifiable rewards (RLVR) has significantly improved the reasoning capabilities of large vision-language models (LVLMs). However, standard on-policy RLVR algorithms face a critical optimization bottleneck in preserving and reinforcing visually grounded reasoning behaviors: valuable visually-grounded reasoning trajectories are discarded after a single update, while uniform token advantage allocation prevents the model from reinforcing critical perception or reasoning steps. To bridge this gap, we propose PIVOT, a dual-level learning framework that anchors policy optimization around informative visual reasoning signals. Specifically, PIVOT introduces a self-calibrated experience replay mechanism, which selectively collects and replays visually-grounded historical experiences as stable reference anchors for policy optimization. Building upon this, we further design a vision-guided advantage allocation mechanism to allocate additional vision-aware advantages to tokens based on their local visual support and impact on downstream reasoning. Extensive experiments across diverse benchmarks demonstrate that PIVOT achieves highly competitive performance in enhancing the multimodal reasoning capabilities of LVLMs.

Xinxin Song, Siyuan Li, Tingxiong Xiao +1

Department of Automation, Tsinghua University · Institute for Brain and Cognitive Science, Tsinghua University

Credit Assignment in RL Experience Replay Reinforcement Learning with Verifiable Rewards Visually Grounded Reasoning RL for VLMs

Sep 15, 2026cs.CV

SAVOR: Self-Aware Visual Grounding via Confidence-Calibrated Reinforcement Learning for Multimodal Hallucination Mitigation

Multimodal large language models (MLLMs) have made strong progress on visual question answering and image captioning, yet they still produce fluent claims about objects, attributes, or relations that are not grounded in the image. Many remedies either modify decoding at test time, which adds latency, or fine tune with preferences such as DPO variants, which teach which answer is preferred but not when the model's own answer is unreliable. We argue that calibrated self assessment is the missing signal. We introduce Savor, a training framework that (i) augments the output schema with token and answer confidence, (ii) optimises the policy with a Group Relative Policy Optimisation (GRPO) objective that penalises calibration error and poor abstention decisions, and (iii) uses the learned confidence at inference time to revisit visual evidence only when the model is uncertain. Experiments on POPE, HallusionBench, AMBER and MMHal-Bench across two recent backbones (InternVL3-8B and Qwen3-VL-8B) show that Savor reduces hallucination while preserving general capability on MME and MMBench, with lower Expected Calibration Error than DPO and decoding baselines.

Zixiu Ding, Zilin Zhao, Yingjie He +3

Central South University · Peking University

Object Hallucination in VLMs Visual Grounding VLM Calibration VLM Hallucination Mitigation RL for VLMs

Sep 14, 2026cs.CV

VideoScout: Learning Agentic Active Exploration with Adaptive Reasoning Pacing for Long Video Understanding

Multimodal Large Language Models (MLLMs) have achieved remarkable progress on short video understanding yet remain limited on long videos due to the limited visual context window. Prevailing approaches rely on uniform frame sampling or recent coarse-to-fine agentic zooming, both of which struggle to localize sparse, decisive evidence in sufficiently long videos. We formulate long video understanding as a \textbf{Sequential Evidence Acquisition (SEA)} problem, in which an agent reads the video turn by turn along the temporal axis, deciding at each turn how fast to watch, what evidence to retain, when to revisit uncertain segments, and when to stop and answer. Inspired by this view, we propose \textbf{VideoScout}, a multi-turn reasoning agent that instantiates the SEA paradigm through adaptive reasoning pacing. Specifically, by dynamically controlling the viewing pace, VideoScout enables efficient traversal of long videos within a bounded visual context window, allowing the agent to access more video content while balancing content analysis depth with reading efficiency. To train VideoScout, we construct VideoScout-66K, a set of over 66K high-quality exploration turns from 10K answer-verified trajectories, and adopt a two-stage pipeline: cold-start supervised fine-tuning teaches the agent per-turn output format, while the Decoupled Clip and Dynamic sAmpling Policy Optimization (DAPO) algorithm performs trajectory-level reinforcement learning with a composite reward that jointly considers answer accuracy, output format compliance, and the temporal alignment between the agent's viewing progress and the teacher's answer timing measured by intersection-over-union (IoU). Extensive experiments on long video understanding and reasoning benchmarks demonstrate that our 7B model achieves strong performance compared with existing trained 7B agentic models.

Weixin Xu, Zhenyu Yang, Bing Wang +2

Video Understanding Long-Video Understanding RL for VLMs Long-Context VLMs

Sep 13, 2026cs.CL

Func-R1: Incentivizing Mathematical Function Reasoning in Multimodal Large Language Models

Performing deliberate mathematical reasoning in visual contexts is a hallmark of advanced Multimodal Large Language Models (MLLMs) and requires a sophisticated synthesis of perceptual grounding and symbolic logic. However, in the realm of mathematical functions, our investigation reveals a critical modality interference phenomenon: even advanced models, while performing textual computational reasoning, tend to disregard or misinterpret essential visual cues. To address this challenge, we propose Func-R1, which synergistically harmonizes precise visual perception and rigorous logical reasoning. Concretely, built upon an explicitly decoupled architecture, we employ a hierarchical post-training framework to progressively identify critical visual evidence and conduct in-depth theoretical reasoning. Furthermore, the Perception-Aligned Theoretic Optimization (PATO) strategy is proposed to steer policy updating towards internalizing fundamental theoretical properties while dynamically rectifying heterogeneous visual information throughout the reasoning process. Extensive experiments across diverse benchmarks demonstrate that Func-R1 delivers the optimal performance among open-source MLLMs, even surpassing GPT-5 with an 8.4% improvement on MathVerse's function-oriented tasks.

Mingze Yin, Xiaohan Wang, Dian Li +8

College of Computer Science and Technology, Zhejiang University · Tencent 3AI Thrust, HKUST (GZ) · Zhongguancun Academy +3

Multimodal Large Language Models Mathematical Reasoning VLM Reasoning RL for VLMs

Sep 7, 2026cs.AI

Eliciting Self-Verification in Multimodal Reasoning Agents with Reinforcement Learning

Reasoning agents increasingly rely on external tools such as web search to answer complex queries. Reinforcement learning (RL) finetuning algorithms such as GRPO have improved long-form reasoning in text-only language models, particularly for coding and mathematics. Reliable tool use in multimodal agents, however, remains challenging because models must interpret text and images while integrating noisy retrieved evidence, often under sparse outcome-level supervision without explicit verification signals. We present Self-Verification via Reinforcement Learning (SVRL), an RL-only finetuning framework that trains multimodal agents to verify and filter retrieved evidence within their own reasoning traces, reducing reliance on external verifiers at inference time. SVRL also introduces a search-aware penalty that discourages unnecessary tool calls and a query-diversity reward that encourages diverse, well-formed search queries, providing fine-grained feedback on when and what to search. Finetuning Qwen-2.5-VL-7B with SVRL on only 5{,}000 visual question answering examples yields consistent gains in multi-hop VQA generalization and tool efficiency across benchmarks. Overall, SVRL narrows the gap between compact agents and much larger proprietary models while requiring substantially lower training and inference cost.

Vishwas Sathish, Viresh Ranjan, Xinliang Zhu +2

University of Washington · Seattle, WA, USA · Amazon +1

Tool Use in VLMs Multimodal QA Multimodal Reasoning LLM Agent Verification RL for VLMs

Sep 7, 2026cs.LG

Stable-MM-R1: Anchoring Multimodal Reasoning Dynamics via Entropy-Guided Stratification

While Reinforcement Learning (RL) effectively incentivizes reasoning in Large Language Models, current pipelines are hindered by training instability and rapid entropy collapse. These limitations often stem from "Rollout Silencing" and low-quality gradient signals in standard sampling procedures. In this work, we propose a robust, data-centric framework to stabilize RL training. We first introduce Potential-Aware Query Mining (PAQM), which filters data dynamically to focus on the "Distillation Zone"---samples with high potential for capability elicitation. Furthermore, we present Hybrid Stratified Replay (HSR), a novel mechanism that restructures batches by stratifying rollouts based on Path Entropy, a rollout-level confidence proxy, and outcome reward. Within each optimization step, HSR reuses current-policy "Stability Anchors" and "Hard Negatives" to construct high-contrast optimization groups, then clears its buffers before the next step. This approach mitigates entropy collapse while improving the utilization of learning signals under limited compute. Our method outperforms strong baselines on complex reasoning tasks, offering a principled solution for stable and efficient RL fine-tuning.

Yimeng Ye, Shuang Chen, Wenxuan Huang +8

Columbia University · CUHK · THU +1

Reinforcement Learning RL Fine-Tuning Sample-Efficient RL RL for VLMs

Sep 3, 2026cs.CV

Unfold The World: Factorize 4D Properties in Reinforcing Spatial Reasoning

Despite the remarkable prowess of Vision-Language Models (VLMs) in general multimodal tasks, they remain fundamentally

flat'' when reasoning about the physical world. We argue that this spatial bottleneck stems from a profound dimensional mismatch: while VLMs are trained to interpret 2D projections, true spatial reasoning demands the recovery of latent 3D geometry and temporal continuity. To conquer this high-dimensional complexity, we advocate a shift from monolithic learning to a

divide and conquer'' paradigm. We present FactoSR, a factorized reinforcement learning framework that explicitly interpret the dimensions collapsed by visual projection. At its core, FactoSR decomposes the monolithic problem of world-consistent reasoning into three orthogonal, geometric sub-objectives: planar correspondence (

XY

), depth consistency (

Z

), and temporal reversibility (

T

). By optimizing these verifiable constraints within a unified policy learning mechanism, we effectively transform an ill-posed projection recovery problem into a series of tangible reasoning steps. Extensive evaluations on multi-view and video benchmarks demonstrate that this elegant decomposition yields substantial gains in 3D and 4D reasoning, achieving a 5.9% boost on VSI-Bench and 4.5% on All-Angles-Bench. Our findings suggest that reinforcing explicit, factorized 4D consistency is a critical step toward evolving VLMs into robust, world-aware reasoners.

Yijun Yang, Shenghe Zheng, Wenbo Li +8

The Hong Kong University of Science and Technology (Guangzhou) · Joy Future Academy · The Hong Kong University of Science and Technology +1

Visual Spatial Reasoning 3D Spatial Reasoning Video Reasoning 4D Scene Understanding Spatial Reasoning RL for VLMs

Sep 1, 2026cs.CV

Controllable Image Captioning with Prompt-Conditioned Scene Rewards

Large Vision-Language Models produce fluent image descriptions but offer limited semantic control: users cannot reliably specify whether captions should emphasize attributes, relations, or particular image regions. We present Fine-grained Captioning Control Using Scene Rewards (FoCUS), a controllable image captioning method that lets users steer captions toward specific semantic emphases through natural-language control prompts. The core idea is a prompt-conditioned control objective based on scene-graph-aligned component scores. Generated captions are parsed and aligned to scene-graph components such as objects, attributes, and relations. These components are differentially weighted, including negative weights, according to the requested emphasis. We optimize this objective with GRPO and further improve its reliability through a stricter object validity threshold and reasoning-based verification for attribute and relation scoring. To evaluate controllability, we introduce Semantic Control and Precision Evaluation (SCoPE), a benchmark with contrastive Include/Avoid constraints for measuring both target content coverage and out-of-scope suppression. Experiments on two VLM backbones show that FoCUS consistently improves controllability and fine-grained caption quality without degrading general caption performance.

Jongyeop Hyun, Taeyoung Kim, Hyounghun Kim

Graduate School of Artificial Intelligence, POSTECH · Department of Computer Science and Engineering, POSTECH

Image Captioning RL for VLMs

Aug 20, 2026cs.CV

Scaffolding Minds: Optimizing Latent Visual Target Representations for Multimodal Reasoning

Latent reasoning has advanced multimodal reasoning through a two-stage training paradigm: (1) a helper image is encoded into latent tokens to teach visual chain-of-thought during a supervised fine-tuning (SFT) stage, and (2) these latent tokens are further refined with reward feedback during a reinforcement learning (RL) stage. In this paper, we identify two key limitations of this framework, one in each stage. First, the SFT stage typically relies on an off-the-shelf vision encoder to encode the helper image, yielding suboptimal latent representations that may not be well aligned with the downstream reasoning task. Second, existing RL methods treat the latent component only through deterministic regularization, which constrains policy drift but does not create alternative latent trajectories for exploration. To address these limitations, we propose Scaffolding Minds. Our approach learns a dedicated scaffolding encoder that provides an optimized target in latent space, and learns both the mean and variance of the RL sampler. We further show that these two improvements are complementary, together yielding substantial gains over strong baselines. Empirically, our method improves over the strongest latent reasoning baseline by +9.5 points on FrozenLake spatial planning, with the gain widening to +19 points on the 32x32 grids, and by +5.6 points on average across nine visual-centric reasoning benchmarks.

Haoqiang Kang, Yinpeng Chen, Luyang Liu +5

Google DeepMind · UC San Diego · Work done at Google DeepMind

Visual Spatial Reasoning Multimodal Reasoning Latent Visual Reasoning RL for VLMs

Aug 13, 2026cs.RO

FIRE-VLA: Failure-Informed Self-Evolution for Vision-Language-Action Models in Autonomous Driving

Reinforcement learning improves autonomous-driving vision-language-action (VLA) models by evaluating trajectories sampled from the current policy. Group relative policy optimization (GRPO) learns from reward differences within each rollout group. When all sampled trajectories are poor, this relative signal can rank failures without identifying behavior outside the failed region. We introduce FIRE-VLA, a failure-informed self-evolution framework that converts such unresolved failures into privileged supervision for the next policy. Low-reward, low-diversity groups trigger self-distillation from a frozen round-start copy of the same model. Teacher and student have the same parameter scale, but only the teacher observes the hidden future trajectory. Supervision follows the student's generated prefix and is restricted to answer tokens, while GRPO remains active for every group. The updated policy supplies the teacher for the next round, allowing the routed failure distribution to change with the policy without requiring a larger external teacher. Starting from the same Qwen2.5-VL-3B SFT checkpoint, the comparison matches student rollout and policy-update counts. On 6,019 examples from 150 held-out nuScenes scenes, FIRE-VLA retains comparable single-sample planning, reduces G=4 mean L2 from 1.848 to 1.500 m, and lowers evaluation-persistent failure prevalence from 13.03% to 11.20%. The reduction in mean error arises mainly from rare severe rollouts rather than uniform improvement across ordinary trajectories.

Hao Dou

Harbin Institute of Technology

VLMs for Autonomous Driving Self-Distillation Autonomous Driving Planning RL for VLMs

Aug 13, 2026cs.RO

Temporal GRPO: Beyond Trajectory-Level Credit in Vision-Language-Action Reinforcement Learning

Outcome-driven reinforcement learning offers a scalable way to post-train vision-language-action (VLA) policies from sparse task-success feedback. In common GRPO-based VLA post-training, one rollout-level advantage is applied to every action in the trajectory. A rollout that completes several valid stages but fails later can therefore penalize the actions that produced its earlier progress. We call this trajectory-level credit aliasing. Temporal GRPO addresses this problem by constructing detectable task stages, aligning each rollout with stage-specific action intervals, and comparing only rollouts that have entered the same stage. The resulting stage advantages are applied to their corresponding intervals in a single policy update. On RoboTwin 2.0, Temporal GRPO improves task success and sample efficiency, with consistent gains across task horizons. Controlled updates on LIBERO-Long preserve shared prerequisite stages and concentrate improvement at the first stage where rollout outcomes diverge.

Yao Zhou, Hang Gao, Fengge Wu +2

Institute of Software Chinese Academy of Sciences, Beijing, China · University of Chinese Academy of Sciences, Beijing, China

Group Relative Policy Optimization Step-Level Credit Assignment in RL Temporal Credit Assignment RL Post-Training Policy Learning RL for VLMs

Aug 12, 2026cs.CV

SCOUT: Unlocking Enhanced Spatial Reasoning via Structured Chain-of-Thought and Multi-Objective Process Reward

Existing Vision-Language Models (VLMs) exhibits a critical bottleneck in robust spatial reasoning. Recent reinforcement learning (RL) methods aim to close this gap with verifiable outcomes, yet they suffer from poor credit assignment across intermediate reasoning steps. Concurrently, structured reasoning approaches overlook the critical depth perception necessary for comprehensive 3D understanding. To address these challenges, we propose SCOUT (Structured Chain-Of-Thought Utilizing Process-Supervised RL Training). Specifically, we design a structured Chain-of-Thought (CoT) framework that explicitly models 3D environmental perception to ensure robust spatial understanding and reasoning. Furthermore, we introduce a novel RL algorithm featuring multi-objective process rewards and a tailored advantage estimation method, facilitating fine-grained credit assignment across distinct segments of the reasoning trajectory. To support our framework, we develop SCOUT-24k, a structured spatial reasoning CoT dataset synthesized through a customized pipeline. Extensive evaluations demonstrate that SCOUT-3B improves upon baseline models by 16.85% and 6.3% on general spatial benchmarks and complex spatial reasoning tasks respectively. Notably, our larger SCOUT-7B even outperforms GPT-4o by a margin of 4.28%. Moreover, despite being trained exclusively on single image, SCOUT-7B exhibits robust out-of-domain generalization to multi-image and video scenarios. These empirical results render SCOUT as a critical step towards next generation of spatially-aware VLMs.

Zile Zhou, Huining Yuan, Weichen Zhang +2

Shenzhen International Graduate School, Tsinghua University, Shenzhen, China

Visual Spatial Reasoning Process Reward Models 3D Spatial Reasoning CoT Reasoning VLM Reasoning RL for VLMs

Aug 11, 2026cs.CV

CARE: Confidence-Aware Reasoning for Reliable Medical VQA

Reinforcement Fine-Tuning (RFT) has enabled medical Multimodal Large Language Models (MLLMs) to produce Chain-of-Thought (CoT) reasoning for visual question answering, yet these models suffer from

\textit{confidence miscalibration}

---a systematic gap between expressed certainty and actual diagnostic accuracy that undermines clinical trust. We propose

\textbf{CARE}

, a

\textbf{C}

onfidence-

\textbf{A}

ware medical

\textbf{RE}

asoning framework that jointly optimizes accuracy and calibration through a dual-stage pipeline. First, a scalable Medical-CoT synthesis provides structured cold-start data for Supervised Fine-Tuning. Second, Group Relative Policy Optimization (GRPO) with a novel

\textbf{Confidence-Aware Reward (CAR)}

mechanism ties the model's confidence to diagnostic correctness within the reward signal. Across three Medical VQA benchmarks,

\textbf{CARE}

achieves the highest diagnostic accuracy while obtaining the lowest Expected Calibration Error and Hallucination Rate, establishing a foundation for trustworthy clinical decision support. Our code is available at https://github.com/anotherbricki/CARE.

Yuetian Du, Yucheng Wang, Zhenyuan Chen +9

Zhejiang University · Ant Group · University of Michigan +1

Medical VQA Medical VLMs VLM Reasoning VLM Calibration VLM Hallucination Mitigation RL for VLMs

Aug 11, 2026cs.CV

MIRA: Medical Image Reflection for Agentic Diagnosis

Medical visual agents can use tools to inspect images and retrieve external knowledge, but indiscriminate tool use may introduce noisy or misleading evidence. Reliable diagnosis therefore requires not only acquiring additional observations, but also verifying whether tool actions are necessary and whether the resulting evidence supports the current hypothesis. We introduce MIRA (Medical Image Reflection for Agentic Diagnosis), a medical visual diagnostic framework for autonomous evidence search and reflective verification. MIRA dynamically invokes image-processing operations, including zooming, grounding, pointing, rotation, and measurement, as well as web search, while evaluating the relevance and consistency of the acquired evidence. We develop MIRA through a two-stage training strategy. First, a tool-augmented Monte Carlo Tree Search data engine explores diverse diagnostic hypotheses and jointly verifies visual grounding accuracy and semantic consistency to construct supervised fine-tuning trajectories. Second, reinforcement learning further improves decision-making through online reflective principle evolution: failure cases are distilled into candidate principles, and only principles that improve held-out rollout rewards are retained. Across nine medical visual reasoning benchmarks, MIRA achieves an average score of 64.73, improving its Qwen3-VL-8B backbone by 7.44 points. It also increases useful tool-use judgments from 56.2% to 73.8% and reduces harmful judgments from 8.9% to 1.6%. Qualitative analyses show that MIRA can re-examine evidence, correct premature conclusions, and adapt its tool-use strategy. Project page: https://MIRA-VL.github.io/

Shengzhi Wang, Jun Yang, Kai Wu +11

Tongji University · École Normale Supérieure – PSL · Nanjing University +2

Medical Image Analysis LLM Agent Verification Tool-Using Vision-Language Agents Medical VLMs RL for VLMs

Aug 11, 2026cs.CV

SafeCap: Improving LVLM Safety with Image Captioning Reinforcement Learning

Large vision-language models (LVLMs) remain vulnerable to jailbreak attacks that exploit visual inputs to bypass safety alignment inherited from their language backbones. We propose SafeCap, a reinforcement-learning framework that aligns LVLMs through learned self-captioning. SafeCap trains a policy model to first generate a safety-relevant image caption and then produce a final answer; the caption is further optimized by whether it enables a frozen LLM to reach a safety-aligned decision. This caption-mediated objective encourages the policy to expose visual cues relevant to safe response generation rather than relying solely on direct refusal supervision. Across five multimodal safety benchmarks and six vision-utility benchmarks, SafeCap substantially improves aggregate safety performance under its intended DirectCap protocol, with gains of 3.7-19.0 points in safety average across four model settings while maintaining comparable or improved vision utility. Under controlled comparisons on matched backbones and data, SafeCap outperforms safety SFT, DPO, and SafeGRPO, demonstrating the effectiveness of caption-mediated reinforcement learning for multimodal safety alignment.

Caoyuan Ma, Wenpu Liu, Weichu Xie +12

The University of Tokyo · Peking University · Wuhan University +4

VLM Robustness Adversarial Attacks on VLMs RL for VLMs

Aug 10, 2026cs.RO

FactorDrive: Adaptive Multi-Step Reasoning Driven by Planning-Critical Factors for End-to-End Autonomous Driving

Vision-language models (VLMs) have advanced scene understanding and enabled explicit reasoning in end-to-end autonomous driving. However, existing methods insufficiently integrate spatial-physical evidence into planning reasoning, while reasoning adaptation remains coarse-grained and falls short of scene-specific planning demands. Furthermore, reasoning-path optimization for higher planning quality remains largely unexplored in autonomous-driving post-training. To address these limitations, we propose FactorDrive, an end-to-end autonomous driving framework for adaptive multi-step reasoning driven by planning-critical factors (PCFs). We first perform large-scale driving-domain instruction tuning to establish foundational driving knowledge. Building on this foundation, we construct PCF-CoT, a chain-of-thought (CoT) dataset that grounds planning reasoning in trajectory-relevant spatial-physical evidence and organizes reasoning around scene-specific PCFs, enabling the composition and depth of reasoning paths to adapt to different planning demands. We further introduce Quality Search-Guided Group Relative Policy Optimization (QS-GRPO), which guides Monte Carlo Tree Search (MCTS) with trajectory-level planning rewards to discover reasoning paths with higher planning quality and uses the resulting responses to optimize the policy through GRPO, thereby improving trajectory planning performance. Extensive experiments on both open-loop (nuScenes) and closed-loop-oriented (NAVSIM) benchmarks demonstrate that FactorDrive achieves state-of-the-art planning performance.

Guolei Huang, Tengfei She, Yuxuan Lu +3

1Southeast University · 2Universiti Malay · 3Ant Group +2

VLMs for Autonomous Driving End-to-End Autonomous Driving Autonomous Driving Planning RL for VLMs

Aug 10, 2026cs.CV

CodecArena: Codec Quality Assessment via Visual Reinforcement Learning

Video coding is advancing into the low and ultra-low bitrate regime, driven by end-to-end codecs that replace the hand-crafted pipeline with jointly optimized neural networks and generative codecs that exploit the priors of video generation models. Yet the dominant metrics, LPIPS and DISTS, measure feature and texture similarity rather than content fidelity: a reconstruction that hallucinates a wrong face or blurs text into convincing strokes can still score well, even when a human rejects it instantly. To address this, we propose CodecArena, the first vision-language framework for video coding quality assessment, casting codec evaluation as source-conditioned comparative reasoning between a reference and its reconstructions. We optimize CodecArena with Facet-GRPO, a visual reinforcement learning scheme that aligns pairwise codec preferences while grounding the verdict in five fidelity facets: identity, objects, text, texture, and temporal consistency. Its facet-anchored reward uses automatically derived facet directions as weak anchors, rather than human per-facet labels, to prevent any single sub-score from dominating the holistic preference and to yield interpretable fine-grained quality judgments. To support training and evaluation in this underexplored regime, we construct two complementary resources: CodecArena-1K, a fully automatic preference dataset of 1,500 comparison groups built from traditional, neural, and generative codec reconstructions with fused vision-language and objective supervision; and CodecArena-Bench, a human-ranked benchmark with source-disjoint videos for fair out-of-domain evaluation. Extensive experiments demonstrate that CodecArena achieves state-of-the-art agreement with human judgments on source-disjoint content across diverse codecs and bitrates, surpassing perceptual metrics and prior vision-language evaluators.

Jiaye Fu, Weiqi Li, Qiankun Gao +5

School of Electronic and Computer Engineering, Peking University · School of Computer Science, Peking University · Pengcheng Laboratory

Video Quality Assessment RL for VLMs

Aug 8, 2026cs.AI

StructReward: Efficient Structured Process Rewards for Self-Correcting Multimodal Reasoning

Reinforcement learning with verifiable rewards (RLVR) has emerged as an effective approach for improving multimodal reasoning. However, most existing methods evaluate an entire response using a binary reward based only on final-answer correctness, thereby discarding the supervision available in intermediate reasoning steps. Process reward models offer finer-grained feedback, but they typically rely on separately trained verifiers, costly chain-of-thought annotations, or online judging by large language models (LLMs). In this work, we introduce StructReward, a compute-efficient framework that provides dense reinforcement signals through structured step-level reward alignment. StructReward represents each generated solution as a sequence of reasoning steps and aligns them with process-labeled reference steps using lightweight numerical, symbolic, and lexical matching rules. The aligned labels are aggregated into a dense process reward and combined with final-answer consistency and output-validity rewards through a gated Group Relative Policy Optimization (GRPO) objective. We further recycle policy rollouts into complementary supervision for response comparison and reflective self-correction, rather than discarding them after policy updates. Separately, we use a strong LLM to rewrite sampled correct trajectories into reflection-oriented training instances, further strengthening the policy's ability to evaluate and refine its reasoning. Since reward computation is performed online without an additional learned verifier or external LLM judge, StructReward substantially reduces the computational overhead of multimodal reinforcement learning. Experimental results show that structured process supervision and rollout recycling provide an efficient path toward self-improving multimodal reasoning.

Yifan Li, Ruxin Sun, Tongzhou Zhao

Process Reward Models Multimodal Reasoning Reinforcement Learning with Verifiable Rewards RL for VLMs

Aug 8, 2026cs.CV

Evidence-RL: Towards Evidence-intensive Visual Reasoning

Vision-Language Models (VLMs) should answer from concrete image evidence rather than language priors, dataset shortcuts, or irrelevant visual context. Existing perception-aware post-training methods encourage image use through global perturbations or attention proxies, but they do not test whether a sampled answer causally depends on the local evidence that supports it. We propose Counterfactual Evidence Disentanglement (CED), a training-time evidence audit for VLM grounding. For each response, CED neutralizes an object-centric Evidence Region and compares the resulting support drop against matched non-evidence Regions. We combine this signal with answer correctness inside GRPO, rewarding correct answers that rely on the evidence path rather than shortcut or nuisance paths. CED uses weak object-level proposals, requires no question-specific evidence annotations, and adds no inference-time overhead. Across nine public benchmarks and four backbones, CED outperforms prior RL-based post-training methods, with targeted analyses verifying its object-centric signal.

Haojie Huang, Xinlei Yu, Chengming Xu +6

National University of Singapore · Zhejiang University · Fudan University +2

Vision-Language Grounding Visually Grounded Reasoning RL for VLMs

Aug 7, 2026cs.RO

TEMPO: Semantic-Action Decoupled RL Post-Training for Vision-Language-Action Models

Vision-language-action (VLA) models are commonly adapted to downstream manipulation tasks via supervised fine-tuning (SFT) or online reinforcement learning (RL) post-training. SFT is prone to distribution mismatch, and existing RL approaches typically apply a single, uniform update strategy to all model components, ignoring their distinct functional roles. We propose TEMPO, a semantic-action decoupled, two-timescale RL post-training framework for VLA models. TEMPO freezes the pretrained vision-language backbone to preserve general semantic representations, and restricts adaptation to two components with dedicated RL optimization loops: the semantic projection layer and the low-level action expert. We update them at different rates--the semantic projection layer infrequently, to keep the latent action stable, and the action expert frequently, to rapidly incorporate control feedback from online interaction. This decoupling RL fine-tuning strategy prevents fast policy updates from destabilizing high-level semantic representations while still allowing the action expert to learn efficiently from online feedback. Experiments on the CALVIN benchmark and real-world manipulation tasks demonstrate that TEMPO consistently outperforms both pretrained state-of-the-art VLA models and the RL post-training baseline, while reaching and maintaining higher evaluation rewards on two real-world tasks.

Ziheng Liu, Quantao Yang

School of Computer Science and Technology, Zhejiang Gongshang University, Hangzhou 310018, China. · Department of Robotics, Perception, and Learning, KTH Royal Institute of Technology, Sweden.

Robotic Manipulation Vision-Language-Action Models RL for VLMs

Aug 7, 2026cs.AI

Gated-BEPO: Confidence-Gated Bellman Credit Assignment for Large Language Model Agents

Training large language model agents in long-horizon environments requires assigning credit from sparse terminal outcomes to individual actions. Existing critic-free methods propagate trajectory-level rewards uniformly across steps, while recent approaches construct step-level groups by matching repeated states and compare actions within each group. The former cannot distinguish useful actions in failed trajectories from ineffective actions in successful ones. The latter rely on step credit derived directly from individual trajectory outcomes and fixed-weight fusion with episode-level credit. We propose Gated-BEPO, which derives step-level credit from empirical rollout graphs. For each rollout group, Gated-BEPO constructs an empirical graph and estimates node values through a mean-backup Bellman fixed point that reflects the empirical action distribution of the current policy. We then accumulate these temporal-difference residuals along each sampled trajectory using generalized advantage estimation, yielding step-level Bellman advantages that capture both immediate and downstream effects. To adaptively fuse episode- and step-level credit, a confidence gate incorporates Bellman credit only at states with multiple observed successors and otherwise uses episode-level credit. Experiments on WebShop, ALFWorld, and visual Sokoban show consistent improvements across language and vision-language models, while diagnostic ablations support the effectiveness of Bellman fixed-point value estimation and show that step-level credit should be incorporated selectively rather than uniformly into the final advantage.

Hongxi Yan, Ziyue Huang, Shichao Fan +1

State Key Laboratory of Virtual Reality Technology and Systems, Beihang University, Beijing, China · Zhongguancun Laboratory, Beijing, China · Qualcomm, Beijing, China

Credit Assignment in RL Step-Level Credit Assignment in RL Temporal-Difference Learning Sparse-Reward RL Policy Learning RL for VLMs+1

Aug 5, 2026cs.CV

Multi-Branch Policy Optimization for Multimodal Large Language Models

Group-based reinforcement learning methods for multimodal large language models typically rely on trajectory-level credit assignment that applies a single advantage to all tokens in a response. However, multimodal reasoning involves substantially higher perceptual uncertainty than text-only settings, where the model must repeatedly re-examine visual information to verify intermediate interpretations, and different visual groundings can lead to divergent reasoning paths, making such uniform credit assignment particularly inadequate and causing relative advantages to progressively degenerate toward zero. To address these challenges, we propose Multi-Branch Policy Optimization (MBPO), a tree-based framework that constructs reasoning trees at vision-language decision boundaries, enabling sibling branches to explore diverse visual hypotheses and assigning segment-level credit through branch-relative advantages. We further introduce a temporal replay buffer to reuse informative segments while controlling policy staleness. Experiments on several multimodal reasoning benchmarks show that MBPO outperforms representative baselines, improving both learning signal quality and optimization efficiency. The code is publicly available at https://github.com/ShuaiLyu0110/MBPO.

Shuai Lyu, Yuning Gong, Ruiling Gao +7

Beijing University of Posts and Telecommunications Beijing, China · Sichuan University Chengdu, China · Shanghai University Shanghai, China +1

Credit Assignment in RL Multimodal Large Language Models Multimodal Reasoning RL for VLMs

Aug 4, 2026cs.CV

CARE-X: Towards Clinically Useful Radiology VLMs with Auxiliary Supervision, Reward-Aligned Learning, and Tool-Augmented Measurement

A clinically useful chest X-ray system must go beyond fluent report generation: it should classify findings with tunable decision thresholds, localize them spatially, and derive the anatomical measurements upon which many diagnoses depend. Today's Vision-Language Models (VLMs) treat these as separate problems, if they address them at all, leaving a gap between what radiologists need and what generative models provide. We introduce CARE-X, a chest X-ray VLM that narrows this gap by unifying auxiliary discriminative supervision with reward-aligned generation. CARE-X augments its generative backbone with focal-loss classification and composite-loss grounding heads, co-trained alongside the language-modeling objective. This auxiliary supervision produces discriminative diagnostic predictions with tunable decision thresholds and precise spatial localization while also improving report quality, providing evidence that structured prediction and generation reinforce one another. Building on this foundation, Decoupled Clip and Dynamic Sampling Policy Optimization (DAPO) leverages task-specific reward signals for report generation, visual question answering (VQA), and spatial grounding, directly optimizing the clinical quality metrics that matter in practice. The result is state-of-the-art performance on the majority of metrics across four report-generation benchmarks, 94.0% VQA accuracy on ReXVQA (+6.0 pp over the next-best baseline), and generative spatial decoding that reaches near parity with dedicated detection heads. Separately, to address measurement-dependent diagnoses, we couple Qwen3-VL-4B-Instruct with native tool-calling capabilities for invoking deterministic measurement tools, while retaining full visual access to the image. This hybrid inference yields +43.6 pp average F1 over perception-only baselines across five measurement-dependent conditions.

Mercy Prasanna Ranjit, Anirban Porya, Sathvik Joel +8

Microsoft Research India · Medha AI, Narayana Health, India. · RTIICS, Narayana Health, India

Radiology Report Generation Chest X-Ray Classification Radiology VLMs Visual Grounding RL for VLMs

Aug 4, 2026cs.RO

Continue or Replan? Bernoulli-Continuation Policy Learning for Adaptive Horizon Execution

Existing chunk-based Vision-Language-Action (VLA) models execute a fixed number of actions (i.e., execution horizon) before replanning, turning replanning into a task-agnostic periodic schedule that is independent of task progress. As a result, when no replanning boundary falls before a critical manipulation stage, it is executed from a stale chunk rather than a freshly replanned one. To address this limitation, we propose Bernoulli-Continuation Policy (BCP), a lightweight, plug-and-play framework for adaptive horizon execution that keeps the base VLA frozen. Given a fixed-length action chunk, its continuation head decomposes execution-horizon selection into a sequence of continue-or-replan decisions, which imposes an ordinal, prefix-sharing inductive bias over candidate horizons rather than treating them as independent classes. Since the optimal horizon for each chunk is not observable, we train this head with reinforcement learning from trajectory-level outcomes and introduce a Replanning-Efficiency Reward that jointly rewards task success and efficient VLA usage, discouraging the policy from collapsing to unnecessarily short horizons. On RoboTwin 2.0 with LingBot-VLA as the base policy, BCP improves the average success rate by +11.08% on 13 low-success tasks and from 89.88% to 93.94% (+4.06%) across all 50 tasks. Although trained only under the Clean setting, BCP generalizes to the Randomized setting, raising the average success rate by +4.06%. It also transfers to a different base policy

π_{0.5}

, achieving a better result on LIBERO (+1.7%) and, notably, on the harder LIBERO-PRO (+6.8%). On a real robot, BCP lifts success from 74% to 92% and from 44% to 84% on two manipulation tasks. Meanwhile, its negligible overhead, combined with higher success, makes BCP's overall runtime even lower than the fixed-horizon baselines.

Weichen Xu, Zhenhua Liu, Lin Luo +8

1Peking University · 2Microsoft Research Asia

Efficient VLA Model Inference Robotic Manipulation Vision-Language-Action Models RL for VLMs

Aug 3, 2026cs.CL

CAVE: Competence-Aware Visual Boundary Evidence Alignment for Video Temporal Grounding

Large vision-language models (LVLMs) have achieved substantial performance gains in Video Temporal Grounding (VTG) through reinforcement learning (RL). However, existing methods primarily rely on outcome correctness rewards that evaluate only the final predicted intervals, leaving boundary-related visual evidence and its correspondence with timestamp predictions insufficiently constrained. In this paper, we delve into timestamp prediction and its underlying boundary-level visual evidence, showing prevalent misalignment between visual evidence and predicted timestamps across widely used benchmarks. To address this issue, we propose Competence-Aware Visual Boundary Evidence Alignment (CAVE), which augments localization optimization with boundary-specific visual evidence rewards to mitigate evidence-timestamp misalignment. Specifically, to explicitly represent the boundary-specific visual evidence, CAVE introduces boundary-specific evidence tokens and initializes their structured generation and distinct boundary semantics through a lightweight supervised warm-up. During RL, the visual boundary evidence alignment reward reinforces the visual attention of special evidence tokens within the ground-truth boundaries, thereby promoting alignment between visual evidence and temporal boundaries. Moreover, performance-aware gating for evidence supervision is designed to adaptively retain evidence guidance for poorly localized groups while reducing it once localization becomes sufficiently accurate to avoid over-constraining fine-grained boundary refinement. Extensive experiments on several public VTG benchmarks demonstrate the effectiveness of our method.

Wei Jia, Zhicong Lu, Yu Chen +5

AMAP, Alibaba Group · University of Chinese Academy of Sciences

Temporal Video Grounding Vision-Language Alignment RL for VLMs

Aug 3, 2026cs.CV

RSVideo: Are Your Vision-Language Models Ready for Remote Sensing Videos?

Remote-sensing videos enable real-time observation of changes in target attributes, short-term activities, and scene evolution. They record motion, actions, interactions, and scene changes that cannot be captured by isolated images. Existing models primarily target single images or discrete temporal observations spanning a long time range. However, a unified evaluation setting for assessing vision-language models on continuous remote-sensing video understanding remains lacking. We introduce RSVideo-10K, a remote-sensing video dataset comprising 10,773 instances, 1.47 million frames, and 17.02 hours of footage, containing both unmanned aerial vehicles and satellite platforms. Its fixed evaluation benchmark, RSVideo-Bench, contains 2,731 test instances and evaluates two complementary aspects of remote-sensing video understanding: L1 Perception and L2 Reasoning, spanning seven capability groups and 17 tasks. Evaluations show that current vision-language models still struggle to recover small local evidence, track short-lived states, and use scene-constrained spatial relations. Based on this analysis, we further propose RSVideo, a reinforcement learning framework for small-target spatiotemporal focusing that selects question-relevant regions across frames and suppresses redundant background tokens. RSVideo achieves a maximum absolute improvement of 9.01% with InternVL3.5-14B and attains the highest accuracy of 40.63% with Qwen3.6-27B across 26 open-source vision-language backbones. Codes will be available at https://github.com/HongjieZhou0329/RSVideo.

Hongjie Zhou, Shiqin Wang, Haoyang Chen +5

Wuhan University · Zhongguancun Academy

Video-Language Model Evaluation Remote Sensing RL for VLMs

Aug 2, 2026cs.CV

Remember-R1: Mitigating Long-Context Visual Forgetting through Reinforcement Learning

Multimodal large language models (MLLMs) increasingly rely on long chain-of-thought reasoning for complex tasks. However, as reasoning sequences lengthen, models may gradually rely less on visual evidence and more on accumulated textual context, leading to visual forgetting. Existing approaches do not directly constrain how visual evidence is used and maintained along the original reasoning trajectory, leaving long-context visual forgetting insufficiently addressed. To address this issue, we propose Remember-R1, a reinforcement learning framework that mitigates long-context visual forgetting by applying process-level supervision directly on the original reasoning trajectory. Specifically, Remember-R1 introduces rewards that encourage broader coverage of matched visual keywords, stronger persistence of visual dependence in later reasoning steps, and greater focus on question-relevant image regions. Experiments across multiple model scales and diverse multimodal benchmarks demonstrate that Remember-R1 consistently improves reasoning performance. Additional analyses further show that it slows the decline of visual attention during generation, supporting its effectiveness in mitigating long-context visual forgetting.

Jianmin Chen, Jiaqi Tang, Wei Wei +9

Hong Kong University of Science and Technology Hong Kong, China · Northwestern Polytechnical University Xi’an, China · Zhejiang University Hangzhou, China

Vision-Language Models Long-Context Modeling VLM Reasoning RL for VLMs